Sliding-window KV inference refers to processing a sequence incrementally while retaining only a fixed-size cache of recent key and value states. It can be applied to pretrained causal transformers at inference time without additional training, while its KV-cache memory remains fixed as more tokens are processed. Because cached states are computed in the context of earlier tokens, they may carry i

---
**📖 中文解读**
以上内容由AI翻译自英文原文，可能存在不准确之处。建议阅读[原文](https://arxiv.org/abs/2609.34049v1)获取最准确的信息。

---
🔗 **原文链接**: [Thinking Outside the Box: Retention and Transmission of Info](https://arxiv.org/abs/2609.34049v1)
🏷️ **转载来源**: ArXiv cs.AI
> 本文由小九AI技术站翻译整理，内容版权归原作者所有。
👤 作者: Timothy DeLise, Seth Cromelin

---
🐾 **小九锐评**

这篇论文来自arXiv预印本，虽然还没有经过同行评审，但选题方向值得关注。
建议先读中文摘要判断是否相关，再看全文细节。

你对这个话题有什么看法？欢迎在评论区讨论 💬

> _转载自 ArXiv cs.AI，内容版权归原作者所有_

---
⏱️ 2026-09-29 13:01